add matrix interface instead of only fomula

[mayrandy]

make

betaboost(y = y, x = x)

possible. Needs to differentiate between fitting mu and phi, and linear and non-linear base-learners.

[mayrandy]

First version works for using a matrix for mu and phi and linear and smooth base-learners.

However, in some settings it might be usefull to incorporate two matrixes, one for mu and one for phi...